Processes `HLSLResourceBindingAttr` attributes that represent
`register(c#)` annotations on default constant buffer declarations and
applies its value to the buffer layout. Any default buffer declarations
without an explicit `register(c#)` annotation are placed after the
elements with explicit layout.
This PR also adds a test case for a `cbuffer` that does not have
`packoffset` on all declarations. Same layout rules apply here as well.
Fixes#126791
Make the name conciser. PPC-specific MCSymbolRefExpr::VariantKind
members will be moved to PPCMCExpr and we will not ue
MCSymbolRefExpr::VariantKind's "generic" members, so there won't be
mix-and-match.
Currently, the spill weight is only determined by isDef/isUse and
block frequency. However, for registers with different register
classes, the costs of spilling them are different.
For example, for `LMUL>1` registers (in which, several physical
registers compound a bigger logical register), the costs are larger
than `LMUL=1` case (in which, there is only one physical register).
To solve this problem, a new target hook `getSpillWeightScaleFactor`
is added. Targets can override the default factor (which is `1.0`)
according to the register class.
For RISC-V, the factors are set to the `RegClassWeight` which is
used to track register pressure. The values of `RegClassWeight`
happen to be the number of register units.
I believe all of the targets with compounded registers can benefit
from this change, but only RISC-V is customized in this patch since
it has widely been agreed to do so. The other targets need more
performance data to go further.
Partially fixes#113489.
Previously, the BufferizableOpInterface implementation for
'tensor.reshape'
listed the 'shape' operand as an alias for the result tensor, causing
unnecessary conflicts with ops that "write" to the shape operand.
The grammar is `!match(str, regex)` and this operator produces 1
if the `str` matches the regular expression `regex`.
The format of `regex` is ERE (Extended POSIX Regular Expressions).
Add a new static analyzer which emits warnings for function call
arguments, local variables, and member variables that are only forward
declared. These forward declaration prevents other WebKit checkers from
checking the safety of code.
It is still better to elide the declaration if possible. To overcome the
false positive undefinedButUsed diagnostic, it seems better to not add
declaration from other units to the set actually.
so that we only need to do it once during recordRelocation. In the
future, we should change fixSymbolsInTLSFixups to apply to MCValue
instead of MCExpr, similar to GNU assembler.
Add a new WebKit checker to validate the correct use of RetainPtr
constructor as well as adoptNS and adoptCF functions. adoptNS and
adoptCf are used for +1 semantics and RetainPtr constructor is used for
+0 semantics.
This fixes compilation issues with GCC and C++23:
```
error: cannot bind non-const lvalue reference of type
'llvm::OptimizationRemarkMissed&' to an rvalue of type
'llvm::OptimizationRemarkMissed'
```
Closes#105778
Improve the check for whether a type can be passed by copy. Currently,
passing by copy is done via the OMP_MAP_LITERAL mapping, which can only
transfer as much data as can be contained in a pointer representation.
I thought I guessed a fix in #130836, but I was wrong.
We actually had the same code in
`flang/cmake/modules/FlangCommon.cmake`.
The check does not pass in flang-rt bootstrap build, because
`-nostdinc++` is added for all `runtimes` checks.
I decided to make the check with the C header, though, I am still
unsure whether it is reliable with a clang that has not been
installed (it is taken from the build structure during flang-rt
configure step).
I verified that this PR enables REAL(16) math entries on aarch64.
Summary:
We used the set only to check if it contains certain metadata nodes.
Replacing the set with a predicate makes the intention clearer and the
API more general.
Test Plan:
ninja check-all
Only print them on TimerGroup destruction (or eagerly when
TimerGroup::printAll() is called).
We should be able to destroy all Timers in a TimerGroup while delaying
printing the stored TimeRecords.
A previous change added the cir-flatten-cfg transform and tested it by
lowering a function with nested scopes to LLVM IR. This change adds
support for invoking the cir-flatten-cfg pass from the cir-opt tool and
adds a new test to verify that functionality in isolation.
Update and generalize materializeBroadcasts to also introduce explicit
broadcasts for VPValues defined in the Plans Entry block.
This fixes a crash when trying to insert the broadcasts generated by
VPTransformState::get after the generating instruction, which isn't
possible after invoke instructions.
Fixes https://github.com/llvm/llvm-project/issues/128838.
If the gather operands in phi nodes are matching and phi nodes may build
up a loop, it may cause a compiler crash with the incorrect def-use
chain. Patch fixes this crash.
This adds support for emitting ClangIR for statements whose value is
ignored. The test case being added (CIR/CodeGen/basic.c) tests a few
more things. The "f1" test case is the only part that's immediately
relevant to this change, but the other cases were part of the same test
in the incubator and they are supported so I brought in the entire test.
When compiling an OpenCL program directly with `clang` using
`--save-temps`, an
error may occur if the program contains OpenCL builtins:
```
test.cl:3:21: error: use of undeclared identifier 'get_global_id'
3 | unsigned int id = get_global_id(0);
| ^
```
This happens because the driver does not add `-fdeclare-opencl-builtins`
when
the input type is `TY_PP_CL`. This PR fixes the issue.
The allocation order of 16 bit registers is vgpr0lo16, vgpr0hi16,
vgpr1lo16, vgpr1hi16, vgpr2lo16.... We prefer (essentially require) that
allocation order, because it uses the minimum number of registers. But
when you have 16 bit data passing between 16 and 32 bit instructions you
get lots of COPY.
This patch teach the compiler that a COPY of a 16-bit value from a 32
bit register to a lo-half 16 bit register is free, to a hi-half 16 bit
register is not.
This might get improved to coalescing with additional cases, and perhaps
as an alternative to the RA hints. For now upstreaming this solution
first.
Instead of having two discrete InputStream and OutputStream helpers,
this merges the two into a unifed 'Transport' handler.
This handler is responsible for reading the DAP message headers, parsing
the resulting JSON and converting the messages into
`lldb_dap::protocol::Message`s for both input and output.
---------
Co-authored-by: Jonas Devlieghere <jonas@devlieghere.com>
Summary:
Currently we dispatch the writing mode off of a runtime enum passed in
by the constructor. This causes very unfortunate codegen for the GPU
targets where we get worst-case codegen because of the unused function
pointer for `sprintf`. Instead, this patch moves all of this to a
template so it can be masked out. This results in no dynamic stack and
uses 60 VGPRs instead of 117. It also compiles about 5x as fast.
This PR creates a new member for task data, which is used to identify
the task in its taskgraph (when ompx taskgraph is enabled).
It aims to remove the overloading of the td_task_id member, which was
used both by the debugger and the taskgraph. This resulted in the
identifier's non-unicity in the case of multiple taskgraphs.
Co-authored-by: Rémy Neveu <rem2007@free.fr>
